CN104977585B - A kind of motion sonar target detection method of robust - Google Patents

Abstract

The present invention relates to a kind of motion sonar target detection method of robust.In one embodiment, methods described includes：By obtaining one group of echo data by sonar battle array and therefrom choosing master data and assistance data with identical reverberation covariance matrix；According to master data, the maximal possibility estimation of the covariance matrix of assistance data computational representation reverberation statistical property；The maximal possibility estimation of target strength is calculated according to master data, is then aided with the nominal guiding vector of target, detection statistic is calculated using the guest's Du test criterion for having robustness to guiding vector mismatch；It whether there is according to detection statistic and the threshold value multilevel iudge target that is determined by false-alarm probability.The present invention to utilize more abundant to echo data by introducing Du's guest's test criterion, so that the robustness of space-time adaptive detector detector under the vectorial mismatch condition of goal orientation greatly improved, and with the CFAR characteristic to unknown Reverberation.

Description

A kind of motion sonar target detection method of robust

Technical field

The present invention relates to the motion sonar target detection method of a kind of object detection method, more particularly to a kind of robust.

Background technology

Marine reverberation is the main interference of shallow sea active sonar, especially when sonar carrier has certain movement velocity When, the reverberation of different azimuth has different Doppler frequency shifts, so that expansion is presented in the reverberation of sonar array element level on frequency spectrum Show as.So the echo signal of low-speed motion will be covered by reverberation, it is impossible to carry out Reverberation Rejection using Doppler, and Also it is difficult to effectively eliminate the reverberation for entering receiver by secondary lobe even if using conventional beamformer.Due to moving the sky of sonar reverberation When coupled characteristic, its target detection problems be sonar worker propose new problem.

Submarine navigation device etc. motion sonar reverberation mechanism of production and part of properties with airborne radar ground clutter very It is similar.1973, Brennan proposed the concept of space-time adaptive processing (STAP) first, and their research shows, STAP energy Enough combine well spatially and temporally handles respective advantage, the platform exercise effect of effective compensation radar, so as to obtain ideal Clutter recognition performance.Subsequent Reed proposes sample covariance matrix and inverted (SMI) detector, so that in theory by STAP Develop into a kind of method for filtering and detecting and organically combine, referred to as space-time adaptive detection (STAD) method.In recent years, STAD Research in motion sonar field is very active, and it is special that result of study shows that STAD methods can make full use of the space-time of reverberation to be distributed Property, improve the detection performance of motion sonar.

For the STAD of point target under Gaussian Profile Reverberation, optimal consistent maximal potential examine be it is non-existent, because This people is based on the solution that the test criterions such as Generalized Likelihood Ratio (GLRT), Rao propose many suboptimums.It is worth noting that These methods are built upon on the basis of two assumed conditions, one assume that can obtain abundance uniform assistance data, be used to The reverberation covariance matrix of to-be-measured cell (master data) is estimated, so as to construct self-adapting detecting statistic.It is auxiliary to ensure uniformity The range cell for helping data typically to be closed on from master data is obtained.Two assume that the direction of target, it is known that the i.e. guiding vector of target It is known, referred to as nominal guiding vector.

In actual applications, the robustness of existing STAD methods has much room for improvement.The actual guiding vector of sonar target and mark Claim guiding vector that mismatch often occurs, its reason includes the deviation of beam position, the under water calibration error of sonar transducer array, many ways Propagate etc..When this mismatch condition occurs, existing method can be by no small Detectability loss.

The content of the invention

The purpose of the present invention is to realize more sufficiently to utilize echo data, greatly improves space-time adaptive detector (STAD) under the vectorial mismatch condition of goal orientation detector robustness, it is and special with the CFAR to unknown Reverberation Property.

To achieve the above object, the embodiments of the invention provide a kind of motion sonar target detection method of robust.It is described Method includes：

One group of echo data is obtained by the echo received by sonar battle array, one chosen in one group of echo data returns Wave number chooses multiple echo datas in one group of echo data in addition to the master data and is used as supplementary number according to as master data According to the master data and the assistance data have identical reverberation covariance matrix；

The main parameter in Du's guest's detection is constituted by the real and imaginary parts of sign target strength parameter, by reverberation covariance Matrix constitutes the auxiliary parameter in Du's guest's detection, and the main parameter is calculated and described using the master data, the assistance data Aid in the maximal possibility estimation of parameter；

Using the estimate and the estimate of the auxiliary parameter of the main parameter, likelihood function under goal hypothesis has been obtained To the value of the difference quotient of main parameter；

According to true value of the main parameter under the maximal possibility estimation under having goal hypothesis and without goal hypothesis and, it is described Difference quotient and the nominal guiding vector obtained by target direction calculating, calculate detection statistic；

By the detection statistic with being compared by giving the obtained threshold value of false-alarm probability, and sentenced according to comparative result The target of breaking whether there is.

It is preferred that, the sample covariance matrix is calculated according to the assistance data and obtained.

It is preferred that, the difference quotient A (θ) can be expressed as：

Wherein z is master data, and Z is that NxN ties up assistance data, θ_ABased on join Amount, K represents to constitute the number of the uniform assistance data of assistance data, H₁Indicate target conditions.

It is preferred that, the threshold value is obtained by setting false-alarm probability using Monte-Carlo Simulation.

It is preferred that, it is described by the detection statistic with being compared by giving the obtained threshold value of false-alarm probability, Ke Yiyou Following formula is determined：

Wherein η is detection threshold, H₀Indicate no target conditions, H₁Target conditions are indicated,^HRepresent conjugate transposition operation,^-1 Matrix inverse operation is represented, v is the nominal guiding vector of target, and S is sample covariance matrix, and z represents master data.

The present invention uses Meng Te-Caro emulation mode in performance detection, is compared with traditional GLRT and Rao detectors Compared with.

The present invention uses Meng Te-Caro emulation mode in performance detection, is compared with traditional GLRT and Rao detectors Compared with.By compare show the inventive method realize to observation data more sufficiently utilize, not only greatly improve STAD in mesh The robustness under guiding vector mismatch condition is marked, and in the case of goal orientation Vectors matching, the inventive method is maintained Extraordinary detection performance.

Brief description of the drawings

Fig. 1 is of the invention in SRR=20dB, traditional GLRT, tradition Rao and tri- kinds of method detectors of Durbin of the present invention Detection probability P_dAnd cos²φ relation curve；

Fig. 2 is of the invention in the case of goal orientation Vectors matching, traditional GLRT, tradition Rao and Durbin of the present invention tri- Plant method detector P_dWith SRR relation curve.

Embodiment

The invention provides a kind of new point target space-time adaptive detection method, combined using dualism hypothesis data Durbin inspection principles, draw the Cleaning Principle formula examined based on Durbin, thus detect that target whether there is.

Implement step as follows：

1) echo data is received based on the linear sonar battle array being made up of N number of array element, thus Point Target Detection can be attributed to as Lower dualism hypothesis：

Wherein H₀And H₁No goal hypothesis is represented respectively and is assumed with the presence of target；N and n_t, t=1 ..., K be it is independent, Zero-mean N-dimensional complex Gaussian reverberation vector, its covariance matrix is E [nn^H]=E [n_tn_t ^H]=M,^HRepresent conjugate transposition operation；z Represent an echo data, also known as master data；z_t, t=1 ..., K represent uniform assistance data of the length for K, and and master data Z has identical reverberation covariance matrix；α is target strength parameter, and its real and imaginary parts are constituted in Du's guest's detection Main parameter；V is to calculate obtained nominal guiding vector by target direction.

For the ease of the design of detector, two simplified styles, i.e., the main parameter vector θ of signal of 2 dimensions are defined_A=[α_R,α_I]^T, Wherein α_R and α_IIt is α real and imaginary parts respectively；N²+ 2 dimensional vectorsWherein θ_BFor N²The nuisance parameter row of dimension Vector, also known as aids in parameter, is made up of covariance matrix M element.Thus, H₁In the case of the master data z and assistance data Z =[=z₁,z₂,...,z_k] joint probability density function be

f(z,Z|θ,H₁)=π^-N(K+1)det(M)^-(K+1)exp{-tr[M^-1((z-αv)(z-αv)^H+S)]} (2)

Wherein S is sample covariance matrix, i.e. S=ZZ^H, det () and tr () represent respectively determinant of a matrix and Trace of a matrix, assistance data Z is that NxN ties up assistance data, and has identical reverberation covariance matrix with master data z.

Observation data are more sufficiently utilized in order to realize, the present invention uses Durbin test criterions

According to main parameter θ_A, auxiliary parameter θ_B, difference quotient A (θ) formation detection statistic with fixation threshold value η be compared, Threshold value controls false alarm rate, and being examined using Durbin to be expressed as：

Wherein, threshold value, θ are obtained using Monte-Carlo Simulation_A,0It is H₀In the case of θ_ATrue value, i.e. θ_A,0=[00]^T； It is H₀In the case of θ_BMaximal possibility estimation； It is H₁In the case of θ_AMaximal possibility estimation, therefore difference quotient A (θ's) is represented by：

Notice

Wherein Re [] and Im [] represent the real and imaginary parts for taking [] interior data respectively.Formula (4) is substituted into formula (3), Draw

Wherein I₂For 2 rank unit matrix,It is H_i, i=0, M maximal possibility estimation in the case of 1.

The present invention is examined by Durbin, takes full advantage of master data z and assistance data Z estimates to calculate M maximum likelihood MeterIt is expressed as：

WillSubstitution formula (5), obtains the nominal guiding vector v of z containing master data, assistance data Z and target A (θ) table Up to formula：

In H₁Have under target conditions, the maximal possibility estimation of target strength parameter alpha is

Formula (7) and formula (8) are substituted into formula (2), the principle type examined based on Durbin is obtained：

Wherein η is one of detection threshold in formula (2) suitably modified.

The principle type examined by Durbin, it can be seen that the observation data of Durbin detectors can be expressed as traditional wide Adopted likelihood probability detection (GLRT) and adaptive matched filter detect the relational expression of (AMF) two kinds of detection statistics：

t_Durbin=t_AMF(1-t_GLRT) (9)

Wherein,

Formula (9) has permanent persistence.Due to (t_AMF,t_GLRT) it is one group of maximal invariant statistic, it is any with this group of invariant The CFAR performance of the detection statistic of composition all, therefore, Durbin detectors proposed by the present invention have assists to unknown reverberation Variance matrix CFAR performance, is easy to the application of actually detected echo signal.

2) performance detection of the invention uses Meng Te-Caro emulation mode, and is carried out with traditional GLRT and Rao detectors Compare.The nominal steering vector v=of target [1 ..., 1]^T/ N, the mixed ratio of letter is defined as SRR=v^HM_- ¹v.The actual guiding vector of target Use v_mRepresent, its mismatch cos between v²φ is weighed, and is specifically defined as

cos²φ=1 represents match condition, i.e. v_m=v.cos²φ<1 represents mismatch condition, and cos²φ values are smaller, v_m Mismatch between v is bigger.

Design parameter in emulation is set to N=8, K=32, false-alarm probability P_fa=10^-3, reverberation is common index phase Close complex Gaussian vector, covariance matrix M=0.9^|i-j|, wherein (i, j) is the coordinate of matrix element.

Fig. 1 is the detection probability P of the present invention, tradition GLRT and tradition tri- kinds of method detectors of Rao in SRR=20dB_d With guiding vector mismatch cos²φ relation curve, it can be seen that when larger mismatch condition occurs in goal orientation vector, The inventive method is maintained to higher detection probability.For example work as cos²During φ=0.4, the P of the inventive method_d=1.0, And tradition GLRT P_d≈ 0.77, traditional Rao P_d<0.05；And work as cos²During φ=0.1, the P of the inventive method_d=0.67, Traditional GLRT P_d<0.06, traditional Rao P_d<0.01。

Fig. 2 is the present invention, tradition GLRT and tradition tri- kinds of method detectors of Rao in the case of goal orientation Vectors matching Detection probability P_dWith the mixed relation curve than SRR of letter, it can be seen that relative to two kinds tradition of the inventive method under match condition The Detectability loss of method is very small, low confidence it is mixed than when Detectability loss within 0.3dB, height letter it is mixed than when the inventive method have Have and the detection performance of tradition GLRT methods quite, hence it is evident that better than traditional Rao methods.

Show that the inventive method is realized by Fig. 1 and Fig. 2 comparative result more sufficiently to utilize observation data, not only Robustness of the STAD under the vectorial mismatch condition of goal orientation is greatly improved, and in the case of goal orientation Vectors matching, this Inventive method maintains extraordinary detection performance again.

Above-described embodiment, has been carried out further to the purpose of the present invention, technical scheme and beneficial effect Describe in detail, should be understood that the embodiment that the foregoing is only the present invention, be not intended to limit the present invention Protection domain, within the spirit and principles of the invention, any modification, equivalent substitution and improvements done etc. all should be included Within protection scope of the present invention.

Claims (5)

1. a kind of motion sonar target detection method of robust, including：

One group of echo data is obtained by the echo received by sonar battle array, a number of echoes in one group of echo data is chosen According to as master data, multiple echo datas in one group of echo data in addition to the master data are chosen as assistance data, The master data and the assistance data have identical reverberation covariance matrix；

The main parameter in Du's guest's detection is constituted by the real and imaginary parts of sign target strength parameter, by reverberation covariance matrix The auxiliary parameter in Du's guest's detection is constituted, the main parameter and the auxiliary are calculated using the master data, the assistance data The maximal possibility estimation of parameter；

Using the estimate and the estimate of the auxiliary parameter of the main parameter, likelihood function has been obtained under goal hypothesis to master The value of the difference quotient of parameter；

According to the main parameter, the auxiliary parameter, the difference quotient and by target direction calculate obtained nominal guiding to Amount, calculates detection statistic；

By the detection statistic with being compared by giving the obtained threshold value of false-alarm probability, and institute is judged according to comparative result Target is stated to whether there is.

2. detection method according to claim 1, it is characterised in that sample covariance matrix is according to the assistance data Calculate what is obtained.

3. detection method according to claim 1, it is characterised in that difference quotient A (θ) is expressed as：

Wherein z is master data, and Z is that N*N ties up assistance data, θ_AFor main parameter, K tables Show the number for the uniform assistance data for constituting assistance data, H₁Indicate target conditions.

4. detection method according to claim 1, it is characterised in that the threshold value is special using covering by setting false-alarm probability Caro emulation is obtained.

5. detection method according to claim 1, it is characterised in that described by the detection statistic and by giving false-alarm The threshold value that probability is obtained is compared, and is determined by following formula：

$t_{Durbin}=\frac{{\left|v^H{S}^{-1}z\right|}^2}{{\left(v^H{S}^{-1}v\right)}^2}\&lsqb;v^H{(s+{\mathrm{zz}}^H)}^{-1}v\&rsqb;\underset{H_0}{\overset{H_1}{\begin{array}{c}>\\ <\end{array}}}\mathrm{\&eta;}$

Wherein η is detection threshold, H₀Indicate no target conditions, H₁Target conditions are indicated, H represents conjugate transposition operation, and -1 represents Matrix inverse operation, v is the nominal guiding vector of target, and S is sample covariance matrix, and z represents master data.